Figure 1.

Weak marginal/strong bivariate signals. The figure displays the scatter plots for 80 observations —with sample sizes n0 = n1 = 40— generated in accordance to the linear, XOR, circular and V-shaped patterns defined in scenarios 1, 2, 3, and 4. The components of the bivariate interaction are weak for class prediction when considered individually; however, they exhibit a high predictive strength when taken jointly.

Arevalillo and Navarro BMC Bioinformatics 2011 12(Suppl 12):S6   doi:10.1186/1471-2105-12-S12-S6